To appear: Int. J. Num. Meth. Eng.

Modelling and Simulation of High-Speed Machining

T. D. Marusich M. Ortiz
Division of Engineering Division of Engineering
Brown University Brown University
Providence RI 02912, USA Providence RI 02912, USA

Abstract
A Lagrangian nite element model of orthogonal high-speed machining is developed. Continuous
remeshing and adaptive meshing are the principal tools which we employ for sidestepping the
diculties associated with deformation-induced element distortion, and for resolving ne-scale
features in the solution. The model accounts for dynamic eects, heat conduction, mesh-on-
mesh contact with friction, and full thermo-mechanical coupling. In addition, a fracture model
has been implemented which allows for arbitrary crack initiation and propagation in the regime
of shear localized chips. The model correctly exhibits the observed transition from continuous
to segmented chips with increasing tool speed.

1 Introduction
The development of a new machining process requires considerable investment of time and
resources (Vasilash [1]). Precise knowledge of the optimal range of the cutting parameters is
essential for a timely startup. Process features such as tool geometry and cutting speed directly
in uence chip morphology, cutting forces, nal product dimensionality and tool life. Computer
simulation of the cutting process can potentially reduce the number of design iterations and
result in a substantial cost savings. Considerable eort has therefore been devoted to the
development of computational models of high-speed machining. Of primary concern is the
determination of the steady-state temperature distribution in the workpiece, tool and chip (e. g.,
Tay et al. [2,3]; Lin et al. [4]). Attempts to model the process of chip formation have for the
most part been based on a predetermined line of separation between the workpiece and chip
(e. g., Strenkowski and Carroll [5]; Komvopoulos and Erpenbeck [6]). Nodes on this line are
separated { and the line `unzipped' { when the tool tip is suciently close, or when a certain level
of plastic strain is attained. Evidently, this simple approach is not capable of predicting surface
roughness and chip morphology. Sekhon and Chenot [7], by contrast, have used mesh adaptivity
to allow for an arbitrary surface of separation. However, elastic strains are not accounted for,
which precludes the computation of residual stresses in the workpiece. In addition, the model
of Sekhon and Chenot [7] does not account for fracture, which severely limits the types of chip
morphologies which can be predicted. It should also be noted that the majority of simulations
conducted in the past idealize the tool as rigid. The tool stiness, however, is known to directly
in uence surface roughness and tool chatter (Kalpakjian, [8]).
In this paper, we present a model of high-speed machining which overcomes some of these
limitations. High-speed machining invariably involves extremely high deformation rates and
unconstrained plastic ow. Lagrangian codes based on a xed mesh have diculties dealing
with this class of problems owing to the large element distortions which, inevitably, are incurred.
Considerable eort has been devoted to the development of elements which perform well when
severely distorted (Pian and Sumihara [9]; Belytschko and Bachrach [10]; Dvorkin and Vassolo
[11]; Simo and Rifai [12]; Belytschko and Bindeman [13]; Simo and Armero [14]; Simo et al. [15]).
A number of alternative formulations have also been proposed for the analysis of unconstrained
ows, including arbitrary Lagrangian-Eulerian (ALE) methods (Hughes et al. [16]; Donea [17];
Liu et al. [18]; Liu et al. [19]; Benson [20]; Ghosh and Kikuchi [21]; Benson [22]) and pseudo-
concentration methods (Thompson [23]; Thompson and Smelser [24]; Dvorkin and Petocz [25]).
However, Lagrangian codes can be used to simulate large unconstrained plastic ows provided
that the solid is continuously remeshed. This is the avenue which we have pursued here. Contin-
uous remeshing and adaptive meshing are the principal tools which we employ for sidestepping
the diculties associated with deformation-induced element distortion, and for resolving ne-
scale features in the solution. We also resort to mesh adaptivity as a means of following the
propagation of cracks along arbitrary paths through the chip. Simulations of high-speed ma-
chining of a structural steel are presented in Section 10. Extensive experimental data is available
for this type of material (Wright and Trent [26]; Ramalingam and Black [27]; Komanduri and
Brown [28], Komanduri et al. [29]). The ability of the model to predict the transition between
continuous and segmented chips with increasing tool speed is particularly noteworthy.

2 Equations of motion
Consider a solid initially occupying a reference conguration B0, and a process of incremental
loading whereby the deformation mapping over B0 changes from n , at time tn , to n+1 = n + u,
at time tn+1 = tn + 4t. Dynamic equilibrium is enforced at time tn+1 weakly by recourse to the
virtual work principle
Z Z Z
B0
Pn+1 : r0 dV0 , (fn+1 , 0an+1 )
 dV0 ,
B0
tn+1
 dS0 = 0
@B0
(1)
where Pn+1 denotes the rst Piola-Kirchho stress eld at time tn+1, fn+1, an+1 and tn+1 are the
corresponding body forces, accelerations and boundary tractions, respectively, 0 is the mass
density on the reference conguration,  is an admissible virtual displacement eld, and r0
denotes the material gradient. Assume for now that a rule has been determined to update the
stress eld of the general form
Pn+1 = P^ (Fn+1 ; state at tn; 4t) (2)
where the deformation gradients
Fn+1 = r0n+1 (3)
are assumed given. Inserting (2) and (3) into (1) one obtains
Z Z Z
^ (r0n+1; state at tn; 4t) : r0 dV0 , (fn+1 , 0an+1 )
 dV0 ,
P tn+1
 dS0 = 0 (4)
B0 B0 @B0
which can be solved for the updated deformation mapping n+1. Upon discretization of (2) with
nite elements the governing equations become
Man+1 + Rint
n+1 = Rn+1
ext (5)
where Z
Miakb = B0
0ik NaNb dV0 (6)
is the mass matrix Z Z
Rext
ia = fiNa dV0 + tiNa dS0 (7)
B0 @B0
is the external force array and Z
Rint
ia = PiJ Na;J dV0 (8)
B0
is the internal force array. In the above expressions, Na, a = 1; : : :; numnp are the shape
functions.
The second-order accurate central dierence scheme is used to discretize (5) in time (Hughes
[35]; Hughes and Belytschko [51]; Belytschko [52]), with the result
dn+1 = dn +
tvn + 12
t2an (9)
 
an+1 = M,1 Rext n+1 , Rn+1
int (10)
vn+1 = vn + 21
t(an+1 + an ) (11)
where d, v and a denote the displacement, velocity and acceleration arrays, respectively. Al-
though the minimum time step used in explicit dynamics is bounded by stability (Hughes [53]),
contact algorithms available for explicit dynamics are more robust and straightforward than
their implicit counterparts. Explicit algorithms are therefore more attractive for problems such
as machining which involve complicated contact situations. Explicit integration is particularly
attractive in three-dimensional calculations, where implicit schemes lead to system matrices
which often exceed the available in-core storage capacity. Yet another advantage of explicit
algorithms is that they are ideally suited for concurrent computing (Mathur et al. [36]).
Explicit schemes require the use of diagonal or lumped mass matrices. No tangent stiness
matrix is assembled and no momentum iterations performed. The bulk of the computational
costs lies in the constitutive calculations at the integration points and in the contact search.
For computational eciency in slow processes, the rate of loading (e. g. cutting tool speed)
can be increased, or the density of the material can be articially increased, which raises the
stability limit (Rebelo et al. [37]). These techniques can be employed when inertia eects are
not signicant.

3 Subcycling
Mesh renement of the type described subsequently can lead to a broad distribution of element
sizes. In explicit calculations, the critical time step
tc for stability scales with the size of the
smallest element, which can result in steps much smaller than required to integrate the coarse
sections of the mesh. Under these circumstances, Belytschko's subcycling algorithm (Belytschko
et al. [38,39]; Belytschko [40]), by permitting each element to be updated according to its own
critical time step, can aord considerable speed-ups. In our implementation of the subcycling
algorithm, element e in the mesh is assigned a time step
te = me
tc, where me is the largest
integer such that
te does not exceed the stability limit of the element. Each node a is then
assigned a time step
ta equal to the largest step
te among all elements connected to it.
Counters are kept for each element and node to determine when an update is necessary. When
an element is updated, displacements at nodes on adjacent slower elements are advanced by
a linear interpolation in time. This strategy has been shown to result in the least noise [39].
Details of the algorithm are given in Box 1.
 i) Initialize counters: ne = me, na = 0
ii) Loop over nodes:
do a = 1; numnp
if na = 1 then
ua = 4tava;n + 12 4t2aaa;n
na = ma
else
na = na , 1
end if
da;n+1 = da;n + ua=ma
end do
iii) Loop over elements:
do e = 1; numel
if ne = 1 then
update element with 4t = 4te
compute internal forces Re;int
n+1
ne = me
else
Re;int
n+1 = 0
ne = ne , 1
end if
end do
iv) Compute velocities and accelerations for cycled nodes:
do a = 1; numnp
if na = ma then 
aa;n+1 = Rexta;n+1 , R a;n+1 =Ma
int
va;n+1 = va;n + 2 4ta [aa;n + aa;n+1 ]
1
end if
end do

Box 1. Subcycling algorithm.

Subcycling is particularly eective for meshes containing elements of greatly disparate sizes.
In the machining applications discussed subsequently, the cutting edge radius sets the scale
for the element size near the tip of the tool, while the mesh size away from the tool typically
scales with the depth of cut. In these applications we have found that subcycling speeds up the
calculations by a factor in the range 2-3.

4 Contact
Machining involves contact between deformable bodies, e. g., the workpiece and the tool. Mesh-
on-mesh contact can also occur between the faces of cracks propagating though the chip and
between detached segments. Here we use the predictor-corrector method developed by Taylor
and Flanagan [49] for the PRONTO2D explicit dynamics code. The bodies in contact can
be deformable or rigid. Two contacting surfaces are designated as master and slave surfaces.
The method starts by calculating the nodal accelerations from the out-of-balance forces and
then computing predictor nodal positions, velocities and accelerations xpred
n+1 , vn+1 and an+1 ,
pred pred
 Slave Surface Slave Surface

M M M
m,k M s,j+1 m,k+1
s,j

Master Surface Master Surface

(a) (b)
Figure 1: a) Predictor conguration of surfaces. b) Kinematically compatible surfaces.
respectively, assuming that no contact has occurred. A resulting predictor conguration wherein
penetration has occurred is shown in Fig. 1a.
In order to establish the contact conditions, an auxiliary consecutive numbering of the nodes
on the contacting surfaces is introduced. The penetration distances s;j for all nodes j on the
slave surface are then determined on the predictor conguration. Here and henceforth, labels
m and s are used to designate the master and slave surfaces, respectively. The contact forces
required to prevent penetration, were the master surface to remain stationary at the predictor
conguration, are given by
Ps;j = Ms;j 2s;j (12)
4t
where Ms;j is the mass of node j on the slave surface. Next, normal acceleration corrections are
introduced which eliminate the unwanted penetration, Fig. 1b. The requisite accelerations are
P
(!s!m;j Ps;j )
corr j
am;k = M + P(! ; (13)
m;k s!m;j Ms;j )
j
X Ps;j
acorr
s;j = m;k ) , M
(!m!s;k acorr (14)
k s;j
where !s!m;j and !m!s;k are weights dependent on position. A Coulomb friction model is also
adopted from Taylor and Flanagan [32]. Let t represent the tangent to the master segment. The
tangential component of the relative predictor velocity between the slave node and the master
segment is given by
pred , X !
4vs;j = t
(vs;j m!s;k vm;k )
pred (15)
k
The force that must be applied to the slave node to cancel its relative tangential velocity, i. e.,
to produce perfect stick, is
Fs;jstick = , Ms;j44tvs;j (16)
The tangential force exerted by the master surface on a slave node cannot exceed the maximum
frictional resistance
F stick
Fs;j = jFs;jstickj min(Ns;j ; jFs;jstickj) (17)
s;j
where Ns;j is the normal contact force given by
s;j
n
Ns;j = Ms;j acorr (18)
where n is the surface normal. The tangential force generates the additional tangential acceler-
ation corrections
acorr Fs;j
s;j = M (19)
s;j
P
!s!m;j Fs;j
j
am;k = , M
corr (20)
m;k
A balanced master-slave approach in which surfaces alternately act as master and slave is em-
ployed however, rigid surfaces are always treated as master surfaces.

5 Thermal eects
In applications such as machining, substantial amounts of heat may be generated due to the
plastic working of the solid and friction at the tool-chip interface. The temperatures attained
can be quite high and have a considerable in uence on the mechanical response. The relevant
balance law, in this case, is the rst law, which can be expressed in weak form as
Z Z Z Z
_
cTdV + hdS = q
rdV + sdV (21)
Bt @Btq Bt Bt
where  is the current mass density, c the heat capacity, T the spatial temperature eld, 
an admissible virtual temperature eld, h the outward heat ux through the surface, q is the
heat ux, s is the distributed heat source density, and Btq the current Neumann boundary.
In machining applications, the main sources of heat are plastic deformation in the bulk and
frictional sliding at the tool-workpiece interface. The rate of heat supply due to the rst is
estimated as
s =  W_ p (22)
where W_ p is the plastic power per unit deformed volume and the Taylor-Quinney [50] coecient
 if of the order of 0.9 (e. g. Kobayashi et al. [41]). The rate at which heat is generated at the
frictional contact, on the other hand is
h = ,t
[ v ] (23)
where t is the contact traction and [ v] is the jump in velocity across the contact. This heat
must be apportioned between the tool and chip. Using transient half-space solutions, the ratio
of the heat supply to the chip, h1, and the tool, h2, can be computed as (cf. Sekhon and Chenot
[7]) p
h1 = pk11c1 (24)
h 2kc 2 2 2
where k,  and c,  = 1; 2, are the thermal conductivity, mass density and heat capacity of
the workpiece and the tool, respectively.
The solid is assumed to obey Fourier's law of heat conduction, however, due to the nite
kinematics care must be exercised in formulating Fourier's law so as to satisfy material frame
indierence. To this end, thermal conductivity is regarded as a property of the crystal lattice
which, therefore, is unchanged by plastic deformation. On this basis, a constant conductivity
tensor D is introduced on the intermediate conguration Bt. The spatial conductivity tensor
then follows as
 eT
D = Fe DF (25)
and Fourier's law can be written in spatial form as
q = ,D
rT (26)
If the lattice conductivity is isotropic, then D = kI and (25) reduces to
D = kBe (27)
where Be is the elastic left Cauchy-Green deformation tensor.
Inserting the nite element interpolation into (21) results in the semi-discrete system of
equations (Belytschko [52])
CT_ + KT = Q (28)
where T is the array of nodal temperatures,
Z
Cab = cNaNbdV (29)
Bt
is the heat capacity matrix, Z
Kab = Dij Na;iNb;j dV (30)
Bt
is the conductivity matrix, and
Z Z
Qa = sNadV , hNadS (31)
Bt @Btq
is the heat source array with h,  = 1; 2 having the appropriate value for the chip or tool
as in (24). However, heat conduction between contacting bodies is not taken into account.
In the applications of interest here, the mechanical equations always set the critical time step
for stability. It therefore suces to integrate the energy equation (28) by the forward Euler
algorithm (Hughes [35]; Hughes and Belytschko [51]; Belytschko [52]), with the result
Tn+1 = Tn + 4tT_ n (32)
CT_ n + KnTn = Qn (33)
We use a lumped capacity matrix C, which elminates the need for any equation solving. Sub-
cycling techniques for parabolic equations (Belytschko et al. [38]) are available which aord
speed-ups comparable to those obtained for hyperbolic problems.

6 Thermo-Mechanical coupling
A staggered procedure (Park and Felippa [54]) is adopted for the purpose of coupling the thermal
and mechanical equations. In our implementation, we allow for dierent meshes for the thermal
and mechanical models, which exchange information by the mesh-transfer operator described in
Section 9. Mechanical and thermal computations are staggered assuming constant temperature
during the mechanical step and constant heat generation during the thermal step. Following
Lemonds and Needleman [31], A mechanical step is taken rst based on the current distribution
of temperatures, and the heat generated is computed from (22) and (23). The heat thus com-
puted is transferred to the thermal mesh and the temperatures are recomputed by recourse to
the forward-Euler algorithm (32) and (33). The resulting temperatures are transferred to the
mechanical mesh and are incorporated into the thermal-softening model described in Section 7,
which completes one time-stepping cycle. A schematic owchart of the staggered procedure is
given in Box 2.

i) Initialize T1 = T0 + 4tT_ 0, n = 0.
ii) Isothermal mechanical step:
fxn; vn; an; Tn+1 g ! fxn+1; vn+1; an+1 ; Tn+1g
iii) Heat generation (bulk + contact)
iv) Rigid conductor step:
fxn+1; vn+1; an+1 ; Tn+1g ! fxn+1; vn+1; an+1 ; Tn+2g
v) n ! n + 1, GOTO (ii)

Box 2. Staggered procedure for thermo-mechanical coupling.

An alternative staggered procedure based on an `adiabatic split' has been proposed by
Armero and Simo [55,56] with a view to ensuring unconditional stability in implicit calcula-
tions.

7 Constitutive model and stress-update algorithm

We adopt a standard formulation of nite deformation plasticity based on multiplicative kine-
matics. In addition, we eect the requisite constitutive updates by a fully implicit update
algorithm proposed by Cuiti~no and Ortiz [43]. The governing constitutive equations and their
discretized counterparts are collected in Box 3.

Constitutive : ! Incremental :
F = Fe Fp ! Fn+1 = Fen+1Fpn+1
F_ pFp,1 = _p R (S ; Q ) ! Fpn+1 = exp (4p R n+1)Fpn
S = S ( 12 log C e ) ! S n+1 = S ( 21 log C en+1 )
Q_ = _p H (S ; Q ) ! Q n+1 = Q n + 4p H n+1
_p = 1 (S ; Q ) ! 4p = 4 t n+1

Box 3. Constitutive framework and state-update algorithm.

We assume that the deformation gradients F admit a multiplicative decomposition into
elastic and plastic parts, Fe and Fp, respectively. The plastic part Fp of the deformation
gradient eld denes an additional (generally incompatible) conguration Bt, or intermediate
conguration. An overbar is used to identify elds dened over the intermediate conguration.
The plastic ow rule, as stated here, determines F_ p fully. The manner in which the ow rule is
discretized is an essential part of the method, and consists of taking the plastic ow direction
R to be constant throughout the increment and equal to its nal value R n+1. This reduces the
ow rule to a system of linear equations for Fp with initial conditions Fpn , the exact solution of
which is given by the exponential mapping.
The the elastic response is expressed in terms of logarithmic elastic strains (1=2) log C e , where
C e = FeT Fe is the elastic right Cauchy-Green deformation tensor. The stress measure S is the
second Piola-Kirchho stress tensor relative to the intermediate conguration. The symbol Q
denotes some suitable set of internal variables referred to the intermediate conguration, with
H the corresponding hardening moduli. Finally, is an eective overstress and  a viscosity
parameter. Note that, by expressing the ow rule, elastic response and hardening laws on the
intermediate conguration, the formulation automatically satises material frame indierence.
The constitutive relations in Box 3 dene a system of nonlinear equations which, for given
Fn+1, can be solved for the updated state variables S n+1 , Q n+1 , F pn+1 , as well as for 4p. As
noted by Cuiti~no and Ortiz [26], it is possible to reduce the system to a single equation for
4p, which can be solved by recourse to a local Newton-Raphson iteration. Cuiti~no and Ortiz
[43] have also noted that the update dened in the foregoing furnishes a material-independent
extension of small-strain updates into the nite deformation range. The full nite deformation
update comprises three steps:
i) Preprocessor step: Compute the predictor logarithmic elastic strains (1=2) log C en+1 from
the given updated deformation gradients Fn+1 . Set the predictor small-strain tensor en+1 equal
to (1=2) log C en+1 . Identify the small-strain stress tensor n with S n and the small-strain internal
variables qn with Q n .
ii) Small-strain update: Eect a small-strain update driven by en+1 with initial conditions n
and qn, to compute n+1 , qn+1 and 4p.
iii) Postprocessor step: Identify S n+1 with n+1 and Q n+1 with qn+1, and compute Fpn+1 by
the exponential mapping.
It should be noted that the steps preceding and following the small-strain update are purely
kinematic in nature, and, hence, material independent. Thus, when it applies, the above pro-
cedure provides a material-independent prescription for extending small-strain updates into the
nite deformation range within the framework of multiplicative plasticity.
The nite plasticity formulation just described has some points of contact with the work of
Weber and Anand [44], such as the use of the exponential mapping for the implicit integration of
the ow rule. Weber and Anand [44] limit their discussion to the isotropic solid and provide a set
of approximate tangents based on a Pade approximation of the logarithmic mapping. Eterovic
and Bathe [45] extended the work of Weber and Anand [44] by considering the case of combined
kinematic-isotropic hardening, but did not addressed the problem of calculating the consistent
tangents, which were given by Cuiti~no and Ortiz [43] in closed form. Finally, it bears emphasis
that neither Weber and Anand [44] nor Eterovic and Bathe [45] discuss the connection between
nite deformation and small-strain constitutive relations and update algorithms.
In a typical high-speed machining event, very high strain rates in excess of 105 s,1 may
be attained within the primary shear zone, while the remainder of the workpiece deforms at
moderate or low strain rates. Under these conditions, a power viscosity law with constant rate
sensitivity m is not adequate. Indeed, the experimental stress-strain rate curves (Klopp et al.
[32]; Clifton and Klopp [33]; Tong et al. [34]) for structural steels exhibit a transition at strain
rates of the order of 105-106 s,1 from low to high rate sensitivity. At low strain rates, a rate
sensitivity exponent in the range 100-200 adequately ts the data, while in the high strain rate
regime a much lower rate sensitivity exponent in the range 5-20 applies. Modelling this type of
behavior is important, since a high rate sensitivity in the primary shear zone may lead to an
elevation in stress which in turn can promote brittle fracture. A simple model which accounts
for this behavior consists of assuming a stepwise variation of the rate sensitivity exponent m
while maintaining continuity of stress. This leads to the relation

_ p! 

!m1
1 + _p = g(p) ; if _p  _t (34)
0

_p! 
_t
!m2 =m1 ,1


!m2
1 + _p 1 + _p = g(p) ; if _p > _t (35)
0 0
where  is the eective Mises stress, g the ow stress, p the accumulated plastic strain, _p0
a reference plastic strain rate, m1 and m2 are low and high strain rate sensitivity exponents,
respectively, and _t is the threshold strain rate which separates the two regimes. In calculations,
we begin by computing _p according to (34), and switch to (35) if the result lies above _t.
Following Lemonds and Needleman [31], we also adopt a power hardening law with linear
thermal softening. This gives
! n1
g = [1 , (T , T0)]0 1 + p
p
(36)
0
where n is the hardening exponent, T the current temperature, T0 a reference temperature, 
a softening coecient, and 0 is the yield stress at T0. It should be noted that, owing to the
staggered integration of the coupled thermal-mechanical equations, the temperature T remains
xed during a mechanical step and, therefore, plays the role of a known parameter during a
stress update.

8 Fracture criteria
The process of segmented and discontinuous chip formation involves the propagation of fractures
through the deforming chip. The simulation of these chip morphologies therefore requires the
formulation of suitable fracture criteria, in conjunction with numerical procedures for nucleating
and propagating a crack through the mesh.
Structural steels can fracture in a brittle or ductile manner (see Ritchie and Thompson
[58] for a review). Brittle fracture, such as occurs below the transition temperature, proceeds
by cleavage. It is generally recognized that, when slip-induced transgranular cleavage is the
dominant mechanism, fracture of mild steels can be described in terms of a critical stress criterion
(Ritchie et al. [57]). In particular, Ritchie et al. [57] found the conditions for mode I brittle
fracture to be consistent with the attainment of a critical opening stress f at a critical distance
l, or RKR criterion. The critical stress f appears to be relatively independent of temperature
and strain rate (Orowan [59]; Knott [60]), and can be inferred from the toughness KIC through
the small-scale yielding relation
f = pKIC (37)
2l
The critical distance l correlates with the spacing of the grain boundary carbides, typically of
the order of two grain diameters (Ritchie et al. [57]).
Under mixed-mode conditions, such as may be expected in machining, the crack may kink
or follow a curved path as it grows. In order to predict the crack trajectory under conditions of
brittle fracture, we adopt the maximum hoop stress criterion of Erdogan and Sih [61]. According
to this theory, the crack propagates along the angle  from the crack face at which the hoop stress
 attains a relative maximum. The resulting trajectory of the crack is such that KII  0 at the
crack tip. Combining the maximum hoop stress and RKR criteria, the condition for mixed-mode
crack growth is taken to be
max  (l; ) = f
 
(38)
with the understanding that the crack propagates at the angle  for which the criterion is met.
Void growth and coalescence is known to be a principal mechanism of ductile fracture in
structural steels [62-65]. Voids are nucleated as a result of fracture or decohesion of carbides
and subsequently grow as the surrounding material strains plastically. The rate of growth of
the voids is accelerated by the blunting of the crack tip, which has the eect of raising the level
of hydrostatic stress at the location of the void. The nal stages of coalescence may be aided
by the development of sheets of smaller voids nucleated from precipitate particles or carbides.
The experimental evidence shows that the fracture toughness of metals depends on the size and
spacing of void nucleating second phase particles [64,66-68].
An approximate analysis of the growth of a spherical void before a blunting crack was given
by Rice and Johnson [69]. The analysis was based on earlier results by Rice and Tracey [70]
pertaining to the growth of an isolated spherical void in an ideally plastic material. Rice and
Johnson [69] used Rice and Tracey's [70] solution with the remote elds identied with the local
stress and deformation elds computed from a slip line solution of a void-free blunting crack.
Rice and Johnson [69] were able to estimate values of the crack tip opening displacement (CTOD)
for fracture initiation which were in good agreement with experiment. Following Ritchie et al.
[71], the critical CTOD criterion for mode I ductile fracture can be recast as the attainment of
a critical value pf of the eective plastic strain at a distance l directly ahead of the crack tip.
Proceeding as before, we express this criterion in the form
max

p(l; ) = pf (39)
with the understanding that the crack propagates at the angle  for which the criterion is met.
Based on Rice and Tracey's solution [70], the critical eective plastic strain can be estimated as
(Ritchie and Thompson [58])
pf  2:48e,1:5p= (40)
where p = kk =3 is the hydrostatic pressure (p > 0 for hydrostatic tension). The strong de-
pendence of the critical eective plastic strain pf on the triaxiality ratio p= is apparent from
(40).

9 Adaptive meshing
One diculty that arises in applying Lagrangian formulations to problems involving uncon-
strained plastic ow, such as machining, is that the mesh may become severely distorted. One
way of sidestepping this diculty, thereby extending the range of applicability of Lagrangian
methods, is to resort to continuous remeshing. In this approach, the connectivity of the -
nite element mesh is redened at regular intervals by triangulating the nodes at their spatial
locations. This process of continuous remeshing, by itself, is capable of eliminating the bulk
of the deformation-induced element distortion. Further gains can be achieved through the use
of simple mesh smoothing algorithms. Besides sidestepping the problem of mesh distortion,
adaptive meshing furnishes a means of simultaneously resolving multiple scales in the solution.
Examples of ne features which drive mesh renement in machining are the mechanical and
thermal boundary layers which develop in the contact region and within localized shear bands.
We also resort to a remeshing technique to propagate cracks through the mesh, as discussed in
Section 10.
The adaptive meshing methodology used here is taken from Ortiz and Quigley [46], with
a number of modications to suit the application addressed here. All meshes are constructed
by Delaunay triangulation (Sloan [42]) and are, therefore, constrained to consist of triangular
elements. The most commonly employed elements of this type are the rst and second-order
isoparametric elements. Mixed elements possessing additional pressure degrees of freedom were
not considered in view of the diculties their use presents in the context of explicit dynamics.
It is well-known that the rst-order simplex suers from volumetric and shear locking which can
result in gross inaccuracies (Nagtegaal et al. [47]; Rebelo et al. [37]). The elements have to be
meshed in a cross-triangle conguration to give acceptable results. However, this conguration
is inappropriate for the unstructured meshes we consider here. By contrast, the second-order
element, which is our element of choice, has higher accuracy and does not lock (Hughes [35]).
The connectivity of the mesh is determined from the set of corner nodes of the elements, with
the midnodes added subsequently. For the applications discussed here, an adaptation criterion
based on the equidistribution of plastic power has proven useful. In this approach, elements
with plastic power contents exceeding a prescribed tolerance TOL, i. e., elements such that
Z

e
W_ pd
> TOL (41)
h

are targetted for renement. Here,

eh denotes the domain of element e, and, for the Mises solid,
the plastic power density is given by
W_ p =  _p (42)
Clearly, this criterion leads to renement in regions of high rate of plastic deformation. The
plastic power equidistribution criterion is amenable to an error minimization interpretation in
some simple model problems (Cuiti~no and Ortiz [48]).
The mesh is adapted at regular intervals by adding new corner nodes at the midsides of
elements targetted for renement. The element connectivity is then completely redened by a
Delaunay triangulation based on the new set of corner nodes. In particular, no hierarchical com-
patibility between subsequent meshes is enforced. Triangulations are eected on the deformed
conguration, which contributes to eliminating much of the mesh distortion introduced by the
ow of material. If, despite this continuous remeshing, elements arise with unacceptable aspect
ratios, the mesh is subjected to laplacian smoothing. In addition, to prevent the deterioration
of the isoparametric mapping, an update of the reference conguration is eected at regular
intervals. Along with adaptive renement, a mesh-coarsening algorithm is used in areas of the
solid which have become inactive. In machining applications, mesh coarsening contributes to
keeping the size of the problem within manageable bounds.
Finally, a transfer operator proposed by Ortiz and Quigley [46] is used to transfer the state
variables between meshes following an adaptation. As shown by Ortiz and Quigley [46], the Hu-
Washizu variational principle provides a prescription for constructing state-transfer operators
once the interpolation of the state variables is dened. We have found it eective to interpolate
the state variables using an auxiliary triangular mesh in which the gauss points of the elements
supply the nodes. This has the eect of smoothing out local spikes in the state elds, such as
inevitably arise near the tip of the tool, which would otherwise corrupt the transferred elds.
In order to preserve the deformation jacobian, logarithmic strains and rotation tensors are
transferred instead of the deformation gradients themselves. Constraints are also applied to the
data to render the deformation and stresses consistent.
Mesh adaption is also our tool of choice for implementing the fracture mechanical aspects
of the model discussed in Section 8. We surround the tip of a growing crack with a rosette of
elements which provide the requisite angular resolution of the near-tip elds, Fig. 2a. The size
of the elements in the rosette is chosen so that a row of gauss points is located at the critical
sampling distance l from the tip. The fracture criteria (39) and (38) are then checked at each
of the critical gauss points. When one of the criteria is met, a new crack segment is placed in
the corresponding direction. This operation entails doubling the node at the previous crack tip,
placing a new node at the new crack tip location, and surrounding the new crack tip with the
 Strain
1
0.93
0.86
0.79
0.72
0.65

Critical distance 0.58

0.51
0.44
0.37
0.3
0.23
0.16
0.09
0.02

Initial conguration (a)

Strain
1
0.93
0.86
0.79
0.72
0.65
0.58
0.51
0.44
0.37
0.3
0.23
0.16
0.09
0.02

Propagated conguration (b)

Figure 2: Procedure for advancing a crack tip.
 (m) (m)
0.0016 0.00150

0.00140

0.0014 0.00130 TEMP

990
0.00120 880
770
0.0012 660
0.00110 550
440
0.00100 330
220
0.0010
110
0.00090
0

0.00080
0.0008
0.00070

0.0006 0.00060

0.00050
0.0060 0.0065 0.0070 0.0075 0.0080 (m) 0.0060 0.0065 0.0070 0.0075 (m)

(m)
(m)

0.0018 0.0018

TEMP
0.0016 0.0016 990
880
770

0.0014 0.0014 660

550
440
330
0.0012 0.0012
220
110
0
0.0010 0.0010

0.0008 0.0008

0.0006 0.0006

0.0060 0.0065 0.0070 0.0075 0.0080 0.0085 (m) 0.0060 0.0065 0.0070 0.0075 0.0080 0.0085 (m)

(m)
(m) TEMP
990
0.0018
880
770
660
0.0016 550
440
0.0015 330
220
0.0014
110
0

0.0012

0.0010 0.0010

0.0008

0.0006

0.0005
0.0060 0.0065 0.0070 0.0075 0.0080 0.0085 (m) 0.006 0.007 0.008 (m)

Mesh (a) Temperature contours (b)

Figure 3: Continuous chip formation.
corresponding rosette of elements, Fig. 2b. By a recursive application of these operations over
successive time steps, brittle or ductile crack growth along arbitrary paths can be simulated.
An alternative method of crack advance based on the introduction of a cohesive law at element
interfaces has been developed by Xu and Needleman [72].

10 Application
The rst application concerns the cutting of a rectangular block of high strength AISI 4340 steel
at a velocity of 30 m/s, with the tool modelled as rigid. The material and process parameters
are collected in Tables 1-3. The thermal properties of the tool are supposed to be matched to
those of the workpiece, giving an equal proportion of frictional heat allotted to the tool and the
chip.
At positive rake angles, the deformation is largely conned to the primary shear zone and to
a boundary layer adjacent to the tool, as expected, Fig. 3. No shear localization occurs and a
continuous chip morphology is predicted. As the chip curls around and contacts the workpiece
a crack initiates on the free surface and propagates through the chip thickness, nally severing
the chip. The extensive mesh renement required to adequately resolve the solution is evident
from Fig. 3. The coarsening of the inactive parts of the mesh is also noteworthy. The maximum
number of nodes in the mesh is near 2600 with about 1100 elements. Time steps were on the
order of 3  10,10 sec which necessitated about 1  106 steps or 20 hours of cpu time on a DEC
3000 workstation.
Average steady state temperatures along the tool-chip interface are near 1000 C and drop
o rapidly into the interior, Fig. 3. Due to the high cutting speed, heat conduction is of
little consequence and conditions are nearly adiabatic. Frictional contact, as opposed to plastic
dissipation in the bulk, accounts for most of the heat generated. The largest accumulated plastic
strains occur within the boundary layer adjacent to the tool. In this region, the ow of material
is facilitated by thermal softening and the plastic strains attain values of the order of 12. Strains
in the chip interior remain within the 1-2 range upon exit from the primary shear zone, and
within 3-5 in the wake of the tool on the workpiece surface.
An inspection of the machined surface reveals surface waviness on the scale of the cutting edge
radius. The material which ows under the tool tip experiences large compressive stresses fol-
lowed by rapid elastic unloading. This unloading may account for some of the surface roughness.
We remark that the Lagrangian character of the simulation and the elastic-plastic constitutive
description adopted enable the computation of the state of residual stress on the cut surface.
Surface roughness and the state of residual stress under the surface may have a signicant eect
of the fatigue life of the component and, consequently, are of engineering interest.
Table 1. Mechanical material constants
E  y m1 m2 _p0 _pt KIC  
200 Gpa 0.3 1090 Mpa 100 5 1 s,1 2105 s,1 30 Mpapm 0.25 0.001 /C

Table 2. Thermal material constants

Density Conductivity Heat capacity
7800 Kg/m 43 W/mC 600 J/Kg C
3

Table 3. Machining simulations

Simulation Speed Rake angle Depth of cut Cutting edge radius
1 30 m/s 10 250 m 25 m
2 10 m/s ,5  500 m 25 m
3 20 m/s ,5 500 m 25 m

Our second application concerns an AISI 4340 steel cut with a negative rake angle at a speed
of 10m=s. The parameters used in the simulation are collected in tables 1-3. Following an initial
transient similar to that observed in the preceding simulation, shear localization occurs along
the primary shear band as the material softens thermally, Fig. 4. The shear band starts out
straight but eventually bows out in a concave down fashion, a process which is accompanied by
the initiation of a ductile crack at the surface of the chip. Fracture proceeds along the band and
is promoted by rapid thermal softening. The crack propagates more than half the way through
the chip thickness before arresting, thus completing the rst segment of a shear-localized chip.
Subsequently, the primary shear band is swept downstream and its temperature prole diuses
out. This is followed by the formation of a new active shear band and the initiation of a ductile
crack at the surface, eventually leading to a second segment. The sequence of events repeats
itself indenitely a results in a `shear localized' chip morphology, in agreement with observation
(Komanduri et al. [29]). In other simulations performed the variation in mesh size, which
directly aects the time step, had little in uence on the location of initiated cracks.
The eective plastic strain attains values in the range 2-3 within the localized zone and 6-12
along the tool face. As in the preceding application, a high degree of mesh renement is required
to adequately resolve the ne features of the solution. It should be noted that the faces of the
cracks which propagate along the primary shear zone often come in contact, which demonstrates
 (m) (m)

0.0030 TEMP
0.0030
810
720
630
540
450
360
0.0025 0.0025 270
180
90
0

0.0020 0.0020

0.0015 0.0015
0.003 0.004 0.005 0.006 (m) 0.003 0.004 0.005 0.006 (m)

(m) (m)

0.0035 0.0035

TEMP
810
0.0030 0.0030 720
630
540
450
360
270
180
0.0025 0.0025 90
0

0.0020 0.0020

0.003 0.004 0.005 0.006 (m) 0.003 0.004 0.005 0.006 (m)

(m) (m)

0.0035 TEMP
0.0035
810
720
630
540
450
360
0.0030 0.0030 270
180
90
0

0.0025 0.0025

0.0020 0.0020

0.003 0.004 0.005 0.006 (m) 0.003 0.004 0.005 0.006 (m)

Mesh (a) Temperature contours (b)

Figure 4: Shear localized chip formation.
the need for mesh-on-mesh contact capability even when the tool is idealized as rigid. The high
plastic strains adjacent to the contact owe partly to the high temperatures which develop in
that region, of the order of 990 C , which result in considerable thermal softening. Temperature
proles within the primary shear band show peaks of 500 C .
The nal example concerns a shear localized chip in which the segments become completely
detached, Fig. 5. The tool geometry is as in the preceding simulation and the cutting velocity is
20 m/s. The initial segment forms as before but is now followed by a crack propagates through
the entire thickness of the chip, causing the chip to become fully detached. Komanduri et al.
[29] have shown that the transitional speed for the onset of segment detachment is near 15m=s
for the material at hand, in agreement with our simulations. The strain rates in the primary
shear band range between 3  105 , 4:5  105 s,1. By contrast, the strain rates which arise in
the same region for a cutting speed of 10 m/s are in the lower range of 1:5  105 , 2  105 s,1.
The elevated stresses which cause a through-thickness crack at the higher speed are furnished
by the variable rate sensitivity model with the threshold strain rate set at 2  105 s,1, which
is in the ballpark of experimental observation (Klopp et al. [32]; Clifton and Klopp [33]; Tong
et al. [34]). Evidently, this threshold separates the strain rate ranges for the cutting velocities
of 10 m/s and 20 m/s considered, ensuring that the transition to segmented chips takes place
between those speeds.

11 Conclusions
We have assembled several numerical techniques which, in combination, enable the simulation
of orthogonal high-speed machining. Finite deformation plasticity with thermal softening, ex-
plicit dynamics, mesh-on-mesh contact with friction, fully coupled heat conduction, continuous
remeshing and fracture are the main elements of the model. Our simulations demonstrate the
ability of the model to predict chip morphologies consistent with observation, perhaps the most
severe test of any machining model. Detailed parametric studies are presently being conducted
and will be reported in a forthcoming publication.

12 Acknowledgements
This work has been funded by the National Science Foundation through Grant DDM-9016568.
The authors are indebted to Steve Wayne of the Valenite Advanced Technology Center for
helpful suggestions and discussions.

13 References
1. Vasilash, G. S. \Cutting Tools," Production Engineering Magazine, (1992) 40-44.
2. Tay, A. O., Stevenson, M. G., de Vahl Davis, G. and Oxley, P. L., \A Numerical Method
for Calculating Temperature Distributions in Machining from Force and Shear Angle Mea-
surements," Int. J. Mach. Tool Des. Res., 16 (1976) 335-349.
3. Tay, A. O., Stevenson, M. G., de Vahl Davis, G., \Using the Finite Element Method
to Determine Temperature Distributions in Orthogonal Machining," Proc. Instn. Mech.
Engrs., 188 (1974) 627-638.
4. Lin, J., Lee, S. L. and Weng, C. I., \Estimation of Cutting Temperature in High Speed
Machining," J. Engrg. Mater. Tech., 114 (1992) 289-295.
 (m) (m)

0.0030 TEMP
0.0030
690
613.333
536.667
460
383.333
306.667
0.0025 0.0025 230
153.333
76.6667
0

0.0020 0.0020

0.0015 0.0015
0.004 0.005 0.006 (m) 0.004 0.005 0.006 (m)
(m)

(m) (m)

0.0035 0.0035

TEMP
810
0.0030 0.0030 720
630
540
450
360
270
180
0.0025 0.0025 90
0

0.0020 0.0020

0.004 0.005 0.006 (m) 0.004 0.005 0.006 (m)

(m) (m)

0.0035 0.0035

TEMP
810
0.0030 0.0030 720
630
540
450
360
270
180
0.0025 0.0025 90
0

0.0020 0.0020

0.004 0.005 0.006 (m) 0.004 0.005 0.006 (m)

Mesh (a) Temperature contours (b)

Figure 5: Discontinuous chip formation.
 5. Strenkowski, J. S. and Carroll, J. T., III, \A Finite Element Model of Orthogonal Metal
Cutting," J. Engrg. Ind., 107 (1985) 349-354.
6. Komvopoulos, K. and Erpenbeck, S. A., \Finite Element Modeling of Orthogonal Metal
Cutting," J. Engrg. Ind., 113 (1991) 253-267.
7. Sekhon, G. S. and Chenot, J. L., \Numerical Simulation of Continuous Chip Formation
During Non-Steady Orthogonal Cutting," Engineering Computations, 10 (1993) 31-48.
8. Kalpakjian, S., \Manufacturing Processes for Engineering Materials" (1984) Addison-
Wesley.
9. Pian, T. H. H. and Sumihara, K., \Rational Approach for Assumed Stress Finite Ele-
ments," Internat. J. Numer. Methods Engrg., 20 (1984) 1685-1695.
10. Belytschko, T. and Bachrach, W. E., \Ecient Implementation of Quadrilaterals with
High Coarse-Mesh Accuracy," Comput. Methods Appl. Mech. Engrg., 54 (1986) 279-301.

11. Dvorkin, E. N. and Vassolo, A. P., \A Quadrilateral 2D Finite Element Based on Mixed
Interpolation of Tensorial Components," Engineering Computations, 6 (1989) 217-224.
12. Simo, J. C. and Rifai, M. S., \A Class of Assumed Strain Methods and the Method of
Incompatible Modes," Internat. J. Numer. Methods Engrg., 29 (1990) 1595-1638.
13. Belytschko, T. and Bindeman, L. P., \Assumed Strain Stabilization of the 4-Node Quadri-
lateral with 1-Point Quadrature for Nonlinear Problems," Comput. Methods Appl. Mech.
Engrg., 88 (1991) 311-340.
14. Simo, J. C. and Armero, F., \Geometrically Nonlinear Enhanced Strain Mixed Methods
and the Method of Incompatible Modes," Internat. J. Numer. Methods Engrg., 33 (1992)
1413-1449.
15. Simo, J. C., Armero, F. and Taylor, R. L., \Improved Versions of Assumed Enhanced
Strain Tri-Linear Elements for 3D Finite Deformation Problems," Comput. Methods Appl.
Mech. Engrg. (to appear).
16. Hughes, T. J. R., Liu, W. K. and Zimmerman, T. K., \Lagrangian-Eulerian Finite Element
Formulation for Incompressible Viscous Flows," Comput. Methods Appl. Mech. Engrg.,
29 (1981) 329-349.
17. Donea, J., \Arbitrary Lagrangian-Eulerian Finite Element Methods," in: T. Belytschko
and T. J. R. Hughes (eds.), Computational Methods for Transient Analysis, North-Holland,
New York (1983) 473-516.
18. Liu, W. K., Belytschko, T. and Chang, H., \An Arbitrary Lagrangian-Eulerian Finite
Element Method for Path-Dependent Materials," Comput. Methods Appl. Mech. Engrg.,
58 (1986) 227-245.
19. Liu, W. K., Chang, H., Chen, J. S. and Belytschko, T., \Arbitrary Lagrangian-Eulerian
Petrov-Galerkin Finite Elements for Nonlinear Continua," Comput. Methods Appl. Mech.
Engrg., 68 (1988) 259-310.
20. Benson, D. J., \An Ecient, Accurate, Simple ALE Method for Nonlinear Finite Element
Programs," Comput. Methods Appl. Mech. Engrg., 72 (1989) 305-350.
21. Ghosh, S. and Kikuchi, N., \An Arbitrary Lagrangian-Eulerian Finite Element Method
for Large Deformation Analysis of Elastic-Viscoplastic Solids," Comput. Methods Appl.
Mech. Engrg., 86 (1991) 127-188.
22. Benson, D. J., \Computational Methods in Lagrangian and Eulerian Hydrocodes, Comput.
Methods Appl. Mech. Engrg., 99 (1992) 235-394.
23. Thompson, E., \Use of the Pseudo-Concentrations to Follow Creeping Viscous Flows dur-
ing Transient Analysis," J. Num. Methods Fluids, 6 (1986) 749-761.
24. Thompson, E. and Smelser, R. E., \Transient Analysis of Forging Operations by the
Pseudo-Concentrations Method," Internat. J. Num. Methods Engrg., 23 (1988) 177-189.
25. Dvorkin, E. N. and Petocz, E. G., \On the Modelling of 2D Metal Forming Processes Using
the Flow Formulation and the Pseudo-Concentration Technique," in: D. R. J. Owen, E.
O~nate and E. Hinton (eds.) Proc. 3rd Int. Conf. Comput. Plasticity, Pineridge Press
Swansea, U. K. (1992) 1037-1051.
26. Wright, P. K. and Trent, E. M., \Metallurgical Appraisal of Wear Mechanisms and Pro-
cesses on High-Speed-Steel Cutting Tools," Metals Technology, (1974) 13-23.
27. Ramalingam, S. and Black, J. T., \An Electron Microscopy Study of Chip Formation,"
Metallurgical Transactions, 4 (1973) 1103-1112.
28. Komanduri, R. and Brown, R. H., \On the Mechanics of Chip Segmentation in Machining,"
J. Eng. Ind, 103 (1981) 33-42.
29. Komanduri, R., Schroeder, T., Hazra, J., von Turkovich, B. F. and Flom, D. G., \ On the
Catastrophic Shear Instability in High-Speed Machining of an AISI 4340 Steel," J. Eng.
Ind., 104 (1982) 121-131
30. Shih, A. J. and Yang, H. T. Y., \Experimental and Finite Element Predictions of Residual
Stresses Due to Othogonal Metal Cutting," Int. J. Num. Meth. Eng., 36 (1993) 1487-
1507.
31. Lemonds, J. and Needleman, A. \Finite Element Analysis of Shear Localization in Rate
and Temperature Dependent Solids," Mechanics of Materials, 5 (1986) 339-361.
32. Klopp, R. W., Clifton, R. J. and Shawki, T. G., \Pressure-Shear Impact and the Dynamic
Viscoplastic Response of Metals," Mechanics of Materials, 4 (1985) 375-385.
33. Clifton, R. J. and Klopp, R. W. \Pressure-Shear Plate Impact Testing," Metals Handbook
Ninth Edition, 8 (1985) 230-239.
34. Tong, W., Clifton, R. J. and Huang, S., \Pressure-Shear Impact Investigation of Strain
rate History Eects in Oxygen-Free High-Conductivity Copper," J. Mech. Phys. Solids,
40 (1992) 1251-1294.
35. Hughes, T. J. R.,\The Finite Element Method," Prentice Hall, (1987)
36. Mathur, K. K., Needleman, A. and Tvergaard, V., \Dynamic 3D Analysis of the Charpy
V-Notch Test," Modelling Simul. Mater. Sci. Eng., 1 (1993) 467-484.
37. Rebelo, N., Nagtegaal, J. C., Taylor, L. M. and Passmann, R., \Industrial Application of
Implicit and Explicit Finite Element Methods to Forming Processes," in AMD-Vol. 156,
Numerical Methods for Simulation of Industrial Metal Forming Processes, ASME, New
York, 1992.
38. Belytschko, T., Smolinski, P., and Liu, W. K., \Stability of Multi-Time Step Partitioned
Integrators for First-Order Finite Element Systems," Comp. Meth. Appl. Mech. Eng., 49
(1985) 281-197.
39. Belytschko, T., Yen, H.-J. and Mullen, R., \Mixed Methods for Time Integration," Comp.
Meth. Appl. Mech. Eng., 17/18 (1979) 259-275.
40. Belytschko, T., \On Computational Methods for Crashworthiness," Comp. Struct. 42 (2)
(1992) 271-279.
41. Kobayashi, S., Oh, S.-I. and Altan, T., \Metal Forming and the Finite Element Method,"
Oxford University Press (1989).
42. Sloan, S. W., \A Fast Algorithm for Constructing Delaunay Triangulations in the Plane,"
Adv. Eng. Software, 9 (1987) 34-55.
43. Cuiti~no, A. M. and Ortiz, M., \A Material-Independent Method for Extending Stress
Update Algorithms from Small-Strain Plasticity to Finite Plasticity with Multiplicative
Kinematics," Engineering Computations, 9 (1992) 437-451.
44. Weber, G. and Anand, L. \Finite Deformation Constitutive Equations and a Time Inte-
gration Procedure for Isotropic Hyperelastic-Viscoelastic Solids," Comput. Meths. Appl.
Mech. Engrg., 79 (1990) 173-202.
45. Eterovic, A. L. and Bathe, K. J., \A Hyperelastic-Based Large Strain Elasto-Plastic Con-
stitutive Formulation with Combined Isotropic-Kinematic Hardening Using the Logarith-
mic Stress and Strain Measures," Int. J. Num. Meth. Eng., 30 (1990) 1099-1114.
46. Ortiz, M. and Quigley, J. J., IV, \Adaptive Mesh Renement in Strain Localization Prob-
lems," Comput. Meth. Appl. Mech. Eng., 90 (1991) 781-804.
47. Nagtegaal, J. C., Parks, D. M. and Rice, J. R., \On Numerically Accurate Finite Element
Solutions in the Fully Plastic Range," Comput. Meth. Appl. Mech. Eng., 4 (1974)
469-484.
48. Cuiti~no, A. M. and Ortiz, M., \Computational Modelling of Single Crystals," Modelling
Simul. Mater. Sci. Eng., 1 (1992) 225-263.
49. Taylor, L. M. and Flanagan, D. P., \PRONTO 2D: A Two-Dimensional Transient Solid
Dynamics Program," Sandia National Laboratories, SAND86-0594, 1987.
50. Taylor, G. I. and Quinney, H., \The Plastic Distortion of Metals," Philos. Trans. Roy.
Soc. London, A230 (1931) 323-362.
51. Hughes, T. J. R. and Belytschko, T., \A Precis of Developments in Computational Meth-
ods for Transient Analysis," J. Appl. Mech., 50 (1983) 1033-1041.
52. Belytschko, T., \An Overview of Semidiscretization and Time Integration Procedures," in:
T. Belytschko and T. J. R. Hughes (eds.), Computational Methods for Transient Analysis,
North-Holland (1983) 1-65.
53. Hughes, T. J. R., \Analysis of Transient Algorithms with Particular Reference to Stability
Behavior," in: Belytschko, T. and Hughes, T. J. R. (eds.), Computational Methods for
Transient Analysis, North-Holland (1983) 67-155.
54. Park, K. C. and Felippa, C. A., \Partitioned Analysis of Coupled Systems," in: T. Be-
lytschko and T. J. R. Hughes (eds.), Computational Methods for Transient Analysis, North-
Holland (1983) 157-219.
55. Armero, F. and Simo, J. C., \Algorithms for Static and Dynamic Multiplicative Plasticity
that Preserve the Classical Return Mapping Schemes of the Innitesimal Theory," Comp.
Meth. Appl. Mech. Engr., 99 (1992) 61.
56. Armero, F. and Simo, J. C., \A-priori Stability Estimates and Unconditionally Stable
Product Formula Algorithms for Nonlinear Coupled Thermoelasticity," Int. J. Plasticity,
9 (1993) 749-782.
57. Ritchie, R. O., Knott, J. F. and Rice, J. R., \On the Relationship Between Critical Tensile
Stress and Fracture Toughness in Mild Steel," J. Mech. Phys. Solids, 21 (1973) 395-410.
58. Ritchie, R. O. and Thompson, A. W., \On Macroscopic and Microscopic Analyses for
Crack Initiation and Crack Growth Toughness in Ductile Alloys," Metall. Trans., 16A
(1985) 233-247.
59. Orowan, E.,\Fracture and Strength of Solids," Rep. Prog. Phys., 12 (1948) 185-232.
60. Knott, J. F.,\Some Eects of Hydrostatic tension on the Fracture Behavior of Mild Steel,"
J. Iron Steel Inst., 204 (1966) 104-111.
61. Erdogan, F. and Sih, G. C., \On the Crack Extension in Plates Under Plane Loading and
Transverse Shear," J. Basic Eng., 85 (1963) 519-527.
62. Rogers, H. C., \Tensile Fracture of Ductile Metals," Trans. Metall. Soc. A.I.M.E., 218
(1960) 498-506.
63. Gurland, J. and Plateau, J., \Mechanism of Ductile Rupture of Metals Containing Inclu-
sions," Trans. Am. Soc. Metals, 56 (1963) 442-455.
64. Cox, T. B. and Low, J. R., \Investigation of the Plastic Fracture of AISI 4340 and 18
nickel-200 Grade Maraging Steels," Metall. Trans., 5 (1974) 1457-1470.
65. Clayton, J. Q. and Knott, J. F., \Observations of Fibrous Fracture Modes in a Prestrained
Low-Alloy Steel," Metal Sci., 10 (1976) 63-71.
66. Pellisier, G. E., \Eect of Microstructure on the Fracture Toughness of Ultra-High Strength
Steels," Engng. Fracture Mech., 1 (1968) 55-75.
67. Green, G. and Knott, J. F., \Initiation and Propagation of Ductile Fracture in Low Steels,"
Trans. ASME 98, Series H, J. Engng Matl Tech., (1976) 37-46.
68. Rawal, S. P. and Gurland, J., \Observations on the Eect of Cementite Particles on the
Fracture Toughness of Spheroidized Carbon Steels," Metall. Trans. A, 8A (1977) 691-698.

69. Rice, J. R. and Johnson, M. A., \The Role of Large Crack Tip Geometry Changes in Planes
Strain Fracture," Inelastic Behavior of Solids ( Kannine, M.F., Adler, W.F., Rosendeld,
A.R. and Jaee, R.I., eds.) 641 McGraw Hill, New York (1970).
70. Rice, J. R. and Tracey, D. M., \On the Ductile Enlargement of Voids in Triaxial Stress
Fields," J. Mech. Phys. Solids, 17 (1969) 201-217.
71. Ritchie, R. O., Server, W. L. and Wullaert, R. A., \Critical Fracture Stress and Fracture
Strain Models for Prediction of Lower and Upper Shelf Toughness in Nuclear Pressure
Vessel Steels," Metall. Trans., 10A (1979) 1557-1570.
72. Xu, X. P. and Needleman, A., \Numerical Simulations of Fast Crack Growth in Brittle
Solids," J. Mech. Phys. Solids (to appear).